def improve_policy(self, **kwargs):
policy_changed = True
i = 0
while policy_changed:
policy_changed = False
for s in self.action_states:
self.env.set_state(s)
best_score = -1
for a in self.env.actions[s]:
reward = self.env.move(a)
next_score = reward + self.discount * self.state_values[
self.env.current_state()]
if next_score > best_score:
best_action = a
best_score = next_score
self.env.undo_move(a)
if best_action != self.policy[s]:
self.policy[s] = best_action
policy_changed = True
self.update_state_value_function(policy='self', **kwargs)
i += 1
if i % 1 == 0:
print("Number of iterations: ", i)
grid_world.print_policy(self.policy, self.env)
grid_world.print_values(self.state_values, self.env)
def run(self, nb_iter=10000, **kwargs):
while self.cur_episode < nb_iter:
self.episode_function(**kwargs)
self.update_explore_threshold()
grid_world.print_policy(self.policy, self.env)
state_values = self.compute_state_values_from_action_values()
grid_world.print_values(state_values, self.env)
def improve_policy(self, nb_iter=1000, **kwargs):
for t in range(nb_iter):
states, actions, rewards = self.perform_episode(**kwargs)
self.update_state_value_function(states, actions, rewards)
self.update_policy()
self.update_explore_threshold(t + 1)
grid_world.print_policy(self.policy, self.env)
grid_world.print_values(self.state_values, self.env)
def improve_policy(self, nb_iter=1000, **kwargs):
for t in range(nb_iter):
states, rewards = self.perform_episode(**kwargs)
state_features = self.compute_features_from_states(states)
self.update_state_value_function(state_features, rewards, t + 1)
self.update_policy()
self.update_explore_threshold(t + 1)
if t % 100 == 0:
print(self.theta)
grid_world.print_policy(self.policy, self.env)
def perform_value_iteration(self, wind, wind_force=0.5):
deltas = [1]
t = 0
while max(deltas) > self.epsilon:
deltas = [None] * len(self.action_states)
for i, s in enumerate(self.action_states):
best_action_value = -1
self.env.set_state(s)
old_value = self.state_values[s]
self.state_values[s] = 0
for a in self.env.actions[s]:
if wind == 'random':
if self.policy[s] == a:
p_a = wind_force
else:
p_a = wind_force * (1 /
(len(self.env.actions[s])) - 1)
elif wind == 'right':
if a == 'R' and a in self.env.actions[s]:
p_a = wind_force
else:
p_a = 0
if self.policy[s] == a:
p_a += (1 - wind_force)
else:
p_a = int(self.policy[s] == a)
reward = self.env.move(a)
next_state = self.env.current_state()
action_value = reward + self.discount * self.state_values[
next_state]
if action_value > best_action_value:
best_action_value = action_value
best_action = a
self.state_values[s] += p_a * action_value
self.env.undo_move(a)
self.policy[s] = best_action
deltas[i] = np.abs(self.state_values[s] - old_value)
t += 1
print("Number of iterations: ", t)
grid_world.print_policy(self.policy, self.env)
grid_world.print_values(self.state_values, self.env)
action_value = reward + self.discount * self.state_values[
next_state]
if action_value > best_action_value:
best_action_value = action_value
best_action = a
self.state_values[s] += p_a * action_value
self.env.undo_move(a)
self.policy[s] = best_action
deltas[i] = np.abs(self.state_values[s] - old_value)
t += 1
print("Number of iterations: ", t)
grid_world.print_policy(self.policy, self.env)
grid_world.print_values(self.state_values, self.env)
if __name__ == "__main__":
optimizer = PolicyOptimizer(environment=grid_world.negative_grid())
grid_world.print_policy(optimizer.policy, optimizer.env)
grid_world.print_values(optimizer.state_values, optimizer.env)
optimizer.perform_value_iteration(wind=None)
# Windy Gridworld: each action has a 50% chance to fail, another action (chosen at random) is performed instead
optimizer = PolicyOptimizer(environment=grid_world.negative_grid())
grid_world.print_policy(optimizer.policy, optimizer.env)
grid_world.print_values(optimizer.state_values, optimizer.env)
optimizer.perform_value_iteration(
wind='right',
wind_force=0.26) # .25 is the threshold to switch optimal agency